This is a preparatory series to make i/o y2038-safe by replacing the use of struct timespec which is not y2038 safe by y2038 safe struct timespec64.
Sockets and userspace interfaces themselves will be changed in a separate series.
Deepa Dinamani (2): select: Use get/put_timespec64 io_getevents: Use timespec64 to represent timeouts
fs/aio.c | 55 ++++++++++++++++++++++++++++++------------------------- fs/select.c | 60 ++++++++++++++++++++++++------------------------------------ 2 files changed, 54 insertions(+), 61 deletions(-)
Usage of these apis and their compat versions makes the syscalls: select family of syscalls and their compat implementations simpler.
This is a preparatory patch to isolate data conversions to struct timespec64 at userspace boundaries. This helps contain the changes needed to transition to new y2038 safe types.
Signed-off-by: Deepa Dinamani deepa.kernel@gmail.com --- fs/select.c | 60 ++++++++++++++++++++++++------------------------------------ 1 file changed, 24 insertions(+), 36 deletions(-)
diff --git a/fs/select.c b/fs/select.c index 9d5f15ed87fe..0cf17fb33cf7 100644 --- a/fs/select.c +++ b/fs/select.c @@ -291,8 +291,7 @@ static int poll_select_copy_remaining(struct timespec64 *end_time, void __user *p, int timeval, int ret) { - struct timespec64 rts64; - struct timespec rts; + struct timespec64 rts; struct timeval rtv;
if (!p) @@ -305,23 +304,22 @@ static int poll_select_copy_remaining(struct timespec64 *end_time, if (!end_time->tv_sec && !end_time->tv_nsec) return ret;
- ktime_get_ts64(&rts64); - rts64 = timespec64_sub(*end_time, rts64); - if (rts64.tv_sec < 0) - rts64.tv_sec = rts64.tv_nsec = 0; + ktime_get_ts64(&rts); + rts = timespec64_sub(*end_time, rts); + if (rts.tv_sec < 0) + rts.tv_sec = rts.tv_nsec = 0;
- rts = timespec64_to_timespec(rts64);
if (timeval) { if (sizeof(rtv) > sizeof(rtv.tv_sec) + sizeof(rtv.tv_usec)) memset(&rtv, 0, sizeof(rtv)); - rtv.tv_sec = rts64.tv_sec; - rtv.tv_usec = rts64.tv_nsec / NSEC_PER_USEC; + rtv.tv_sec = rts.tv_sec; + rtv.tv_usec = rts.tv_nsec / NSEC_PER_USEC;
if (!copy_to_user(p, &rtv, sizeof(rtv))) return ret;
- } else if (!copy_to_user(p, &rts, sizeof(rts))) + } else if (!put_timespec64(&rts, p)) return ret;
/* @@ -704,17 +702,15 @@ static long do_pselect(int n, fd_set __user *inp, fd_set __user *outp, const sigset_t __user *sigmask, size_t sigsetsize) { sigset_t ksigmask, sigsaved; - struct timespec ts; - struct timespec64 ts64, end_time, *to = NULL; + struct timespec64 ts, end_time, *to = NULL; int ret;
if (tsp) { - if (copy_from_user(&ts, tsp, sizeof(ts))) + if (get_timespec64(&ts, tsp)) return -EFAULT; - ts64 = timespec_to_timespec64(ts);
to = &end_time; - if (poll_select_set_timeout(to, ts64.tv_sec, ts64.tv_nsec)) + if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec)) return -EINVAL; }
@@ -1051,12 +1047,11 @@ SYSCALL_DEFINE5(ppoll, struct pollfd __user *, ufds, unsigned int, nfds, size_t, sigsetsize) { sigset_t ksigmask, sigsaved; - struct timespec ts; - struct timespec64 end_time, *to = NULL; + struct timespec64 ts, end_time, *to = NULL; int ret;
if (tsp) { - if (copy_from_user(&ts, tsp, sizeof(ts))) + if (get_timespec64(&ts, tsp)) return -EFAULT;
to = &end_time; @@ -1102,10 +1097,10 @@ SYSCALL_DEFINE5(ppoll, struct pollfd __user *, ufds, unsigned int, nfds, #define __COMPAT_NFDBITS (8 * sizeof(compat_ulong_t))
static -int compat_poll_select_copy_remaining(struct timespec *end_time, void __user *p, +int compat_poll_select_copy_remaining(struct timespec64 *end_time, void __user *p, int timeval, int ret) { - struct timespec ts; + struct timespec64 ts;
if (!p) return ret; @@ -1117,8 +1112,8 @@ int compat_poll_select_copy_remaining(struct timespec *end_time, void __user *p, if (!end_time->tv_sec && !end_time->tv_nsec) return ret;
- ktime_get_ts(&ts); - ts = timespec_sub(*end_time, ts); + ktime_get_ts64(&ts); + ts = timespec64_sub(*end_time, ts); if (ts.tv_sec < 0) ts.tv_sec = ts.tv_nsec = 0;
@@ -1131,12 +1126,7 @@ int compat_poll_select_copy_remaining(struct timespec *end_time, void __user *p, if (!copy_to_user(p, &rtv, sizeof(rtv))) return ret; } else { - struct compat_timespec rts; - - rts.tv_sec = ts.tv_sec; - rts.tv_nsec = ts.tv_nsec; - - if (!copy_to_user(p, &rts, sizeof(rts))) + if (!compat_put_timespec64(&ts, p)) return ret; } /* @@ -1198,7 +1188,7 @@ int compat_set_fd_set(unsigned long nr, compat_ulong_t __user *ufdset, */ static int compat_core_sys_select(int n, compat_ulong_t __user *inp, compat_ulong_t __user *outp, compat_ulong_t __user *exp, - struct timespec *end_time) + struct timespec64 *end_time) { fd_set_bits fds; void *bits; @@ -1271,7 +1261,7 @@ COMPAT_SYSCALL_DEFINE5(select, int, n, compat_ulong_t __user *, inp, compat_ulong_t __user *, outp, compat_ulong_t __user *, exp, struct compat_timeval __user *, tvp) { - struct timespec end_time, *to = NULL; + struct timespec64 end_time, *to = NULL; struct compat_timeval tv; int ret;
@@ -1317,12 +1307,11 @@ static long do_compat_pselect(int n, compat_ulong_t __user *inp, { compat_sigset_t ss32; sigset_t ksigmask, sigsaved; - struct compat_timespec ts; - struct timespec end_time, *to = NULL; + struct timespec64 ts, end_time, *to = NULL; int ret;
if (tsp) { - if (copy_from_user(&ts, tsp, sizeof(ts))) + if (compat_get_timespec64(&ts, tsp)) return -EFAULT;
to = &end_time; @@ -1386,12 +1375,11 @@ COMPAT_SYSCALL_DEFINE5(ppoll, struct pollfd __user *, ufds, { compat_sigset_t ss32; sigset_t ksigmask, sigsaved; - struct compat_timespec ts; - struct timespec end_time, *to = NULL; + struct timespec64 ts, end_time, *to = NULL; int ret;
if (tsp) { - if (copy_from_user(&ts, tsp, sizeof(ts))) + if (compat_get_timespec64(&ts, tsp)) return -EFAULT;
to = &end_time;
On Sat, Aug 5, 2017 at 6:12 AM, Deepa Dinamani deepa.kernel@gmail.com wrote:
Usage of these apis and their compat versions makes the syscalls: select family of syscalls and their compat implementations simpler.
This is a preparatory patch to isolate data conversions to struct timespec64 at userspace boundaries. This helps contain the changes needed to transition to new y2038 safe types.
Signed-off-by: Deepa Dinamani deepa.kernel@gmail.com
fs/select.c | 60 ++++++++++++++++++++++++------------------------------------ 1 file changed, 24 insertions(+), 36 deletions(-)
Looks all good to me,
Reviewed-by: Arnd Bergmann arnd@arndb.de
struct timespec is not y2038 safe. Use y2038 safe struct timespec64 to represent timeouts. The system call interface itself will be changed as part of different series.
Timeouts will not really need more than 32 bits. But, replacing these with timespec64 helps verification of a y2038 safe kernel by getting rid of timespec internally.
Signed-off-by: Deepa Dinamani deepa.kernel@gmail.com Cc: linux-aio@kvack.org --- fs/aio.c | 55 ++++++++++++++++++++++++++++++------------------------- 1 file changed, 30 insertions(+), 25 deletions(-)
diff --git a/fs/aio.c b/fs/aio.c index 8f0127526299..7ca6b7a00368 100644 --- a/fs/aio.c +++ b/fs/aio.c @@ -1289,20 +1289,10 @@ static bool aio_read_events(struct kioctx *ctx, long min_nr, long nr,
static long read_events(struct kioctx *ctx, long min_nr, long nr, struct io_event __user *event, - struct timespec __user *timeout) + ktime_t until) { - ktime_t until = KTIME_MAX; long ret = 0;
- if (timeout) { - struct timespec ts; - - if (unlikely(copy_from_user(&ts, timeout, sizeof(ts)))) - return -EFAULT; - - until = timespec_to_ktime(ts); - } - /* * Note that aio_read_events() is being called as the conditional - i.e. * we're calling it after prepare_to_wait() has set task state to @@ -1824,6 +1814,25 @@ SYSCALL_DEFINE3(io_cancel, aio_context_t, ctx_id, struct iocb __user *, iocb, return ret; }
+static long do_io_getevents(aio_context_t ctx_id, + long min_nr, + long nr, + struct io_event __user *events, + struct timespec64 *ts) +{ + ktime_t until = ts ? timespec64_to_ktime(*ts) : KTIME_MAX; + struct kioctx *ioctx = lookup_ioctx(ctx_id); + long ret = -EINVAL; + + if (likely(ioctx)) { + if (likely(min_nr <= nr && min_nr >= 0)) + ret = read_events(ioctx, min_nr, nr, events, until); + percpu_ref_put(&ioctx->users); + } + + return ret; +} + /* io_getevents: * Attempts to read at least min_nr events and up to nr events from * the completion queue for the aio_context specified by ctx_id. If @@ -1842,15 +1851,14 @@ SYSCALL_DEFINE5(io_getevents, aio_context_t, ctx_id, struct io_event __user *, events, struct timespec __user *, timeout) { - struct kioctx *ioctx = lookup_ioctx(ctx_id); - long ret = -EINVAL; + struct timespec64 ts;
- if (likely(ioctx)) { - if (likely(min_nr <= nr && min_nr >= 0)) - ret = read_events(ioctx, min_nr, nr, events, timeout); - percpu_ref_put(&ioctx->users); + if (timeout) { + if (unlikely(get_timespec64(&ts, timeout))) + return -EFAULT; } - return ret; + + return do_io_getevents(ctx_id, min_nr, nr, events, timeout ? &ts : NULL); }
#ifdef CONFIG_COMPAT @@ -1860,17 +1868,14 @@ COMPAT_SYSCALL_DEFINE5(io_getevents, compat_aio_context_t, ctx_id, struct io_event __user *, events, struct compat_timespec __user *, timeout) { - struct timespec t; - struct timespec __user *ut = NULL; + struct timespec64 t;
if (timeout) { - if (compat_get_timespec(&t, timeout)) + if (compat_get_timespec64(&t, timeout)) return -EFAULT;
- ut = compat_alloc_user_space(sizeof(*ut)); - if (copy_to_user(ut, &t, sizeof(t))) - return -EFAULT; } - return sys_io_getevents(ctx_id, min_nr, nr, events, ut); + + return do_io_getevents(ctx_id, min_nr, nr, events, timeout ? &t : NULL); } #endif
On Sat, Aug 5, 2017 at 6:12 AM, Deepa Dinamani deepa.kernel@gmail.com wrote:
struct timespec is not y2038 safe. Use y2038 safe struct timespec64 to represent timeouts. The system call interface itself will be changed as part of different series.
Timeouts will not really need more than 32 bits. But, replacing these with timespec64 helps verification of a y2038 safe kernel by getting rid of timespec internally.
Signed-off-by: Deepa Dinamani deepa.kernel@gmail.com Cc: linux-aio@kvack.org
Nice cleanup of the compat path!
static long read_events(struct kioctx *ctx, long min_nr, long nr, struct io_event __user *event,
struct timespec __user *timeout)
ktime_t until){
ktime_t until = KTIME_MAX; long ret = 0;if (timeout) {struct timespec ts;if (unlikely(copy_from_user(&ts, timeout, sizeof(ts))))return -EFAULT;until = timespec_to_ktime(ts);}/* * Note that aio_read_events() is being called as the conditional - i.e. * we're calling it after prepare_to_wait() has set task state to@@ -1824,6 +1814,25 @@ SYSCALL_DEFINE3(io_cancel, aio_context_t, ctx_id, struct iocb __user *, iocb, return ret; }
+static long do_io_getevents(aio_context_t ctx_id,
long min_nr,long nr,struct io_event __user *events,struct timespec64 *ts)+{
ktime_t until = ts ? timespec64_to_ktime(*ts) : KTIME_MAX;struct kioctx *ioctx = lookup_ioctx(ctx_id);long ret = -EINVAL;if (likely(ioctx)) {if (likely(min_nr <= nr && min_nr >= 0))ret = read_events(ioctx, min_nr, nr, events, until);percpu_ref_put(&ioctx->users);}return ret;+}
I guess these two functions are small enough that they could be merged into one, saving a few lines. Then again, fs/aio.c seems to generally use fairly short functions doing not too much at once, so your approach maybe fits better with the style of the subsystem.
Either way,
Reviewed-by: Arnd Bergmann arnd@arndb.de
Arnd
I guess these two functions are small enough that they could be merged into one, saving a few lines. Then again, fs/aio.c seems to generally use fairly short functions doing not too much at once, so your approach maybe fits better with the style of the subsystem.
I don't see a problem with combining the two functions either. Unless someone has a strong preference, I will leave it the way it is currently handled.
Thanks, Deepa
-
Arnd Bergmann -
Deepa Dinamani